Not applicable.
Not applicable.
The present invention generally relates to image processing carried out in connection with medical imaging, such as digital radiography. More particularly, the present invention relates to a method and apparatus for identifying non-clinical regions in a medical image and correcting therefor. In addition, the present invention relates to a method and apparatus for enhancing imaging processing of edges of medical diagnostic images.
In the past, radiation imaging systems have been proposed which capture images representative of an X-ray scan of a region of interest obtained from a patient. The scan of the region of interest is captured upon a film, screen, digital detector or the like and later converted by a radiation image processing system to a radiation image comprised of gray scale pixels representative of the region of interest of the patient displayable to a physician. Before display, the radiation image is image processed, such as through noise filtering, contrast enhancement, data compression, and the like.
According to one image processing technique, a digital signal representation of a radiation image is first decomposed into a multi-resolution (MR) representation containing localized image detail at multiple scales or frequencies. During MR processing, the image is decomposed into a series of processed images at multiple resolution levels. Multi-resolution imaging decomposes an original or basic image into different resolution or frequency images/bands. The decomposed image also contains a residual image at an even lower resolution level. The images at each level may be referred to as the levels of a pyramid. The differences in frequency for the images of each pyramid level illustrate various image features at different resolutions. Once the decomposition function is completed, the images formed at the various levels of the pyramid are modified in a desired manner, such as to perform edge or contrast enhancement and the like. The images at the various levels of the pyramid may also be combined through weighted functions to afford each pyramid level image a different amount of impact upon a resulting image. The final processed image is computed through a reconstruction algorithm. For instance, edge enhancement may be achieved by heavily weighting the first few decomposed images (i.e., pyramid level zero).
Many different algorithms have been proposed for performing multi-resolution imaging. A common characteristic of multi-resolution imaging algorithms is that decomposition is achieved by applying a filter to the pixel values of the image. For instance, a low pass filter may be used in a convolution process to create each pyramid level. During each iteration through the multi-resolution processing technique, the low pass filter operates upon each pixel within the image, including all of the image border pixels. The filter is applied to the image border by overlaying a filter kernel upon each pixel along the image border and modifying the image pixel based upon the surrounding pixel values overlaid by the filter kernel. For instance, an image pixel overlaid by the center element of a filter kernel may be replaced with an average of the sum of the products of the filter kernel elements and overlaid image pixels. When a filter operates upon the border pixels of the image, a portion of the filter kernel extends beyond the edge of the image. The filter kernel elements that extend beyond the image edge still contribute to the filtered value that replaces an image pixel. Hence, an error is introduced into the resulting filtered image pixel since a portion of the filter kernel extends beyond the image edge. Errors formed by the filter kernel when processing an image edge appear as an artifact in the image to be displayed ultimately.
When using multi-resolution algorithms to process images, such as during digital radiography, incorrect treatment of edges and edge regions may create image artifacts which extend far into the interior of the image. The edges may be caused by collimators located about the patient region of interest during the X-ray process. Edges may also be caused by defective pixels along the border of the X-ray detector. Also, the edge of the radiation field also creates artifacts when processed.
During multi-resolution (MR) processing, the images are decomposed into successive lower resolution images (or pyramid levels) via the convolution of the radiation image with a low pass filter of finite size (e.g., a 3xc3x973, 5xc3x975, 7xc3x977 element array, etc.). Each subsequent pyramid level is typically one-half the size of its predecessor level. Hence, it is preferable that the original image size be integer divisible by 2N, where N is the number of levels to be computed during decomposition. Each pyramid level contains specific frequency content. After the transformation of each pyramid level, an output image is constructed by reversing the decomposition process.
One problem that occurs during decomposition is, when the filter kernel is centered over pixels at the edge of the radiation image, no image data exists for convolution with the outer elements of the filter kernel extending beyond the image. For instance, when the filter kernel is centered over a corner pixel of a radiation image and the filter kernel is a 3xc3x973 array, five elements of the filter kernel have no underlying image data, upon which to operate. In the past, it has been proposed to xe2x80x9cpadxe2x80x9d the radiation image with a border of zeros, where the width of the border was dependent upon the size of the filter kernel. Decomposition is a recursive algorithm and thus the padding must be iteratively placed around the radiation image at each pyramid level. For instance, if the MR algorithm uses eight pyramid levels, the image at each of the eight pyramid levels must be padded with zeros. The padding does not accurately reflect the image data values along the image edge and thus creates an artifact during decomposition in each pyramid level image. The artifacts are carried to each lower level pyramid image and magnified at each pyramid level when a new border of zero padding is added until reaching the bottom of the pyramid where the error becomes quite large.
X-ray systems have recently been proposed which utilize digital X-ray detectors that offer much improved resolution. Digital detectors have in turn enabled X-ray systems to greatly enhance illustrations of small image features. Heretofore, conventional digital detectors inherently exhibited enough noise to mask the artifacts or errors caused by MR processing of image borders. The inherent detector noise covered up artifacts created during the image processing of the borders. Modern digital detectors now offer higher signal-to-noise ratios and thus artifacts created during MR processing of an image border have become more noticeable.
X-ray systems utilize image detectors having a generally fixed size. However, it is not always necessary to view a region of a patient as large as the detector. The size of the patient region that is exposed to X-rays is reduced by blocking a portion of the X-ray source from the patient with a collimator. By way of example, an X-ray detector may capture a radiation image formed of a 2Kxc3x972K array of pixel values. The radiation image may include a region associated with (e.g. located behind) the collimator. The border between the collimator and the patient""s area of interest is defined by an edge since the collimator blocks radiation, while the patient""s body passed a majority of the radiation. The edge has a broad frequency signal component.
An operator may collimate the field of interest to expose a smaller region of interest than the full field of view. When a collimator is used, a region exists within the image which is termed a non-clinical region. Non-clinical regions may be removed or xe2x80x9ccroppedxe2x80x9d such that the resulting image no longer has a desired shape or size, such as sides that are integer divisible by 2N. The gray scale transition of image data from inside the field of view to outside the field of view contains substantial frequency content. Consequently, multi-resolution algorithms typically perform edge enhancement of high frequencies and dynamic range compression of the lower frequencies. Thus, conventional multi-resolution techniques enhance the xe2x80x9cnon-clinicalxe2x80x9d edges and subtract the artificial low frequency information across the entire image. In the past, the artificial low frequency components of the radiation image formed by the collimator may influence interior pixels as far as 10-20 cms into an image.
Further, unattenuated non-clinical regions of the image may influence interior-clinical regions. The potential influence of unattenuated non-clinical regions upon interior regions is due to the fact that the radiation image has a very broad band frequency spectrum. Past systems have not satisfactorily limited the influence of unattenuated regions upon clinical regions.
A need exists for an improved image processing method and apparatus that overcomes the above discussed problems. It is an object of the preferred embodiments of the present invention to meet this need.
A method and apparatus are provided according to a preferred embodiment of the present invention for processing a radiation image containing clinical and non-clinical regions. The non-clinical regions are identified and removed or cropped to form an intermediate preprocessed image conforming to the boundaries of the clinical region. The intermediate preprocessed image is combined with a border region along at least one side of the clinical region of the radiation image. The size and configuration of the newly added border are controlled in order to achieve a resultant image having a predetermined size and configuration easily processed in a multi-resolution image processing technique to be performed upon the radiation image. Once the border region is formed and combined with the clinical region, a multi-resolution imaging technique is performed to image process the clinical region and the border. By xe2x80x9ccroppingxe2x80x9d the non-clinical region and substituting therefore a border of desired size and shape, the method and apparatus of the preferred embodiment pretreat a radiation image to minimize artifacts and errors associated with multi-resolution processing of non-clinical regions and edges.
According to an alternative embodiment, the border region is assigned pixel values which are determined as a function of the pixel values in the clinical region. Optionally, the pixel values of the border region may mirror adjacent pixel values in the clinical region.
According to yet a further alternative embodiment, a method and apparatus are provided in which the X-ray system utilizes predetermined information and/or detection algorithms to locate the non-clinical regions (e.g., identify the position of a collimator, or defective detector pixels and the like). The non-clinical regions are cropped such that all remaining image pixels correspond to a field of interest from a patient. The perimeter of the image is padded with a border having a width sufficient to result in a preprocessed image having dimensions that are integer divisible by 2N. The pixel values for the border are computed based on a low pass mirroring function. During pyramid decomposition, the decomposed image at each pyramid level is iteratively padded with a border of pixels that is assigned values based on a low pass mirroring function. The size of the border is chosen to form resultant images at each pyramid level of a size suitable for use with an MR algorithm. For instance, the border may have a width of the (filter size xe2x88x921)/2. Once the MR pyramid is built, a transform function is performed and an output image is reconstructed and recropped to form an image having a size corresponding to the original radiation image.
In yet a further alternative embodiment, a method and apparatus are provided for an MR mirroring technique. Once a non-clinical region is identified and removed, the remaining clinical region is combined with a border to enlarge the image to a desired target size. Temporarily, the pixel values associated with the border may be assigned zero values. Next, the MR mirroring technique may scan the newly formed image along at least some horizontal rows of pixels. The MR mirroring technique copies the values of pixels from along the row inside of the clinical region to corresponding pixel values outside the clinical region in the border. The mirroring technique may be repeated on both the left and right sides of the image. Next, the MR mirroring technique scans vertically along at least some columns of the pixels in the newly formed image. The MR mirroring technique copies values of pixels from inside the clinical region of the image to corresponding pixel locations in each column outside of the clinical area. The vertical scanning technique may be performed on both the top and bottom sides of the image. Entire rows and columns need not be scanned. Instead, only a number of pixels within the clinical area need be scanned sufficient to assign pixel values to the border regions. The pixel values in the border regions may be further altered by performing a convolution operation with a low pass filter to remove any discontinuities.
Optionally, the radiation image and border may be formed of an array of pixel values and the multi-resolution imaging technique may use a filter kernel to modify pixel values of the clinical region based on pixel values of the border that correspond to adjacent pixel values of clinical region. The non-clinical region may be located around or inside the clinical region. When the non-clinical region is formed inside the clinical region, the non-clinical region may be overlaid with a border configured as a patch and containing pixel values assigned based on the clinical region pixel values surrounding the non-clinical region. The border may be along one side of the clinical region or more.
A further embodiment of the present invention includes a method and apparatus of preprocessing a portion of a radiation image corresponding to a region of interest representative of a patient. A field of view is collimated to expose a smaller patient region of interest than a full field of view of the radiation image. A radiation image is obtained which contains a region of interest and a collimated image region bordering the region of interest along at least one side. The collimated image region is identified, such as by identifying the position of the collimators through the use of sensors, based on predetermined collimator locations, based on an image recognition technique which processes the radiation image to identify the collimated image region and the like. The collimated image region, once identified, is replaced with corrective pixel value data that is determined based on data values from the region of interest.